This is a question that’s been lingering throughout the year. The 2025 Berlin Marathon was one of the first races in the qualifying period for the 2027 Boston Marathon. The weather was unseasonably warm and number of BQ’s at the race was much lower than usual.
Certainly that means fewer applicants. But just how big of an impact will it have?
The trouble is that Berlin, like London, Tokyo, and Sydney, has a lower conversion rate of qualifiers to applicants than the American Majors and other American races. So losing 4,000 BQ’s from Berlin isn’t the same as losing 4,000 BQ’s from Chicago.
In the next two weeks, the qualifying period will wrap up and it’ll be time to write up a final prediction. So I want to spend a little time tying up these loose ends. Today, we’ll focus on the Berlin question and think about how to quantify how much of an impact it’ll have.
What If You Exclude Berlin?
Let’s start with a high level question. What does the data look like if you just exclude Berlin?
Currently, with races through August 23, there are 11.42% more finishers and 2.33% more qualifiers than last year. Using the basic assumption that the number of applicants rises and falls with the number of qualifiers, that would increase the number of applicants slightly (~34,000) and produce a slightly higher cutoff.
If you set aside the Berlin results, though, there’s a different story. The number of finishers has grown even more quickly (+14.20%) and the number of qualifiers is up far more (+11.09%). If this was the basis for the overall change, the number of applicants would surge from ~33,000 to almost 37,000.
The actual outcome should be somewhere in the middle – 35,000 or 36,000 applicants.
What If We Work From the Number of Entrants?
When BAA announces the cutoff, they typically report the five races that yielded the most entrants. Last year, Berlin was third on that list with 1,127. That means 23,235 entrants came from other races.
Let’s break that apart and estimate how much each component would change.
The number of qualifiers from this year’s Berlin race (4,749) was 54% of what it was the previous year (8,781). Applying that rate to the number of entrants from last year (1,127) yields an estimated 610 entrants for the 2027 Boston Marathon.
The number of BQ’s at other races has increased 11.09%, so we’ll apply that increase to the remaining number of entrants from last year (23,235). This yields 25,812 entrants for the 2027 Boston Marathon.
Add those two groups together, and you’re looking at approximately 26,422.
If the number of accepted applicants is similar to last year (24,362) this means cutting about 2,000 additional applicants. The past two years, there have been ~1,800 applicants per minute with a buffer of ~BQ-5. So that means a solid minute and change over last year’s cut-off.
That would put things around 5:45, give or take. Higher than the straight projection based on the full results set (currently 5:19).
What If We Account for the Distribution of Buffers?
Something else I’ve been thinking about lately is that the distribution of buffers has changed and the number of applicants could change as well due to the lottery.
The first is obvious when you look at the breakdown of the current qualifier data in the tracker. So far this year, there are 16,249 qualifiers with a 20+ minute buffer, compared to 15,204 last year. The 10-20 minute group is also much larger and the 5-10 minute group is slightly larger. This is offset by a decrease in the 0-5 minute group.
If this changes the distribution of buffers among the actual applicants, it could cause an issue with the linear regression that predicts the cutoff time from the number of rejected applicants.
The flip side of this is that the lottery could change incentives for runners. There may be (at least a slight) increase in the share of runners with very small buffers who apply. This, too, would alter the fundamentals of that linear regression.
Another way to think about this, then, is to estimate how many applicants come from each group of qualifiers and then count from the bottom up.
I started by projecting out the total number of qualifiers we should expect after the last couple weeks are complete. The current number will increase by about 1.34%, based on the results from last year.
Then, I calculated the effective conversion rate of qualifiers in each group last year. Combining those rates with the projections and applying it to the data from the 2026 qualifying period would yield 32,888 applicants. So it’s slightly underestimating the actual outcome from last year (33,249).
Applying this to this year’s data yields an expected outcome of 33,326 applicants. So a slight increase over last year but not huge.
However, if we focus just on runners with 10 minute buffers, there would have been 14,214 applicants last year and 15,282 applicants this year.
Adding in the 5-10 group, we’re up to 22,991 last year and 24,245 this year. Assuming this is a slight undercount that means the group of runners with 5+ minute buffers will more than fill up field and the cutoff will be over five minutes.
Combine This With the Berlin Data
Finally, let’s combine this idea of calculating the estimated yield of applicants with the data from Berlin and the rest of the races.
When you isolate the results from Berlin, this formula heavily overestimates the number of expected applicants. This is because a) it’s based on the overall conversion rate (and it’s lower at Berlin) and b) the conversion rate is inflated because it also accounts for races that aren’t in the dataset.
But using the same model and applying it to the 2026 results from Berlin project 3,246 applicants with a buffer of 5:00 or more. This is about three times the actual number (1,127 applicants with a buffer of 4:34 or larger).
Applying this model to the 2027 data yields 1,562 project applicants. That’s a decrease of 48%. So similar to the example above, we’d expect the actual number of applicants from Berlin to be reduced by about one half – yielding 600 or so applicants.
When you isolate the rest of the results and remove Berlin, the model now underestimates things. Last year’s data would predict 19,745 applicants from other races – compared to the actual outcome of 23,235.
But applying the projection to this year’s data suggests 22,684 applicants. That’s an increase of 14.9%. Applied to last year’s total – 23,235 – that yields 26,697 applicants.
When you add in the 600 applicants from Berlin, that’s a total of ~27,300 applicants. That’s about 3,000 more than last year’s field size.
This would point towards a worst case scenario of ~6:00.
The Bottom Line on Berlin and the Cutoff
We’ve looked at this from a few different angles, and they all paint a similar picture: a big drop in BQ’s from Berlin is obscuring the picture. It’s dragging down the projection, resulting in a current projection of 5:19.
The number of BQ’s from other races has increased much more dramatically. And this will likely far outpace the 500 or 600 entrant decrease from Berlin. The net result will be 2,000 to 3,000 who need to get cut.
A more conservative bet would be ~5:45. In this case, the results from Berlin are dragging things down by ~20-25 seconds. A worst case scenario would be ~6:00. In that case, the impact of Berlin would be more like a 40 second underestimate.
At the end of the day, this confirms what I’ve been suggesting all year – that the results from Berlin are artificially lowering the projection somewhat. And it provides a little context for how much the estimate is likely underestimating things.
More to come over the next two weeks as I put the final pieces together for a final prediction.